Board Ties, Learning, and Emerging Market Entry 413 an incoming interlock, and in the second stage

Board Ties, Learning, and Emerging Market Entry 413 an incoming interlock, and in the second stage

estimated our main results controlling for the hazard.

To identify the hazard of forming incoming ties properly in the first stage, we included five instruments that affected the likelihood of having incoming ties but not the probability of entry in the second stage. We reasoned that certain attributes of firms’ headquarters (HQ) city would partially explain the ability of firms to attract incoming directors but be unrelated to the choice to invest in Eastern Europe. The five instruments were the population of the HQ city, the number of cities within 50 kilometers of HQ, whether the HQ city had a first division (Bundesliga) soccer team, the yearly precipitation in the HQ region, and whether the HQ were located in a state that provided economic assistance to other German states. Since treatment regression requires the endogenous variable to be dichotomous, we created an indicator coded as 1 if the firm had one or more incoming ties and 0 otherwise. To be able to compare the effect of incoming ties to other types of ties, we also dichotomized outgoing and

indirect ties 2 . The results are shown in Table 4. In Models 9–11, we treat incoming ties (regard- less of whether they are formed by CEOs or not) as potentially endogenous. Model 9 demonstrates that, after accounting for selection, the pattern of tie influence remains as predicted by H1a and H1b. Models 10 and 11 compare effects across firms with high and low experience in other East- ern European countries. As in H3a–H3c, indirect ties increase in influence whereas incoming and outgoing ties decrease in influence as firms gain their own experience (p < 0.05). In Model 12 we treat incoming CEO ties as potentially endoge- nous. Our reasoning is that firms would be espe- cially desirous to invite CEOs of other firms to their boards to benefit from the status and expe- rience of these individuals. We continue to find support for H2b (p < 0.01). Our hypotheses con- tinue to be supported after accounting for the selection of incoming ties—providing further evi- dence that learning is the mechanism explaining these results.

2 We also attempted to estimate our models treating incoming interlocks as a continuous outcome in the first stage but found

that identification was extremely weak or impossible because of the high prevalence of firms with no incoming interlocks to directors with experience in particular countries.

Other tests The high correlation between incoming and indi-

rect interlocks (see Table 2) raises a concern that multicollinearity may affect our results. We took two steps to address this concern. First, we esti- mated VIF scores for each variable and for the overall model (Allison, 1984). None of the scores approached problematic levels. Second, we ran our analysis excluding indirect interlocks, and our find- ings regarding incoming and outgoing interlocks were unaltered. Thus, the correlation between incoming and indirect interlocks is not biasing the results.

We also assessed the robustness of our results to alternative event history estimation procedures including the rare event procedure suggested by King and Zeng (2001), the Cox (1972) propor- tional hazard model, and piecewise exponential regression (Lee, 2007). In all cases, our findings remained robust.